Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Claude Effort Test Builds a 1.03M-Triangle Blender Scene

AI summary

A Juejin AI analysis of the open-source realcodebear/blender-fable-5.2 project compared Claude Fable 5.2 running Blender 5.2 in Medium Effort and Max Effort modes. The reported Max Effort run used three agents and generated 1,028,114 triangles across three procedural scenes, with estimated scene-generation cost rising from $6.22 to $113.96, or $118.84 including the main session. The article says the scenes used no external models or image textures and were baked into GLB files for a Three.js viewer.

In-depth insight

The repository analysis offers concrete evidence for AI-driven 3D engineering, showing how agent effort, repeated visual inspection, headless rendering and context-heavy iteration can substantially increase geometry and cost.

零外部模型、零外部图片贴图,同一模型(Claude Fable 5.2)用纯 Python 脚本在 Blender 5.2 中从零生成 3 个 3D 场景。 当投入成本从 $6.22 拉满到 $113.96(18 倍),代码从 1,439 行膨胀到 8,740 行,总三角面数从 12.6 万跃升至 102.8 万——这背后到底发生了什么?

最近 GitHub 上的开源项目 realcodebear/blender-fable-5.2 做了一组非常有意思的对照实验:让同一个 AI 模型在 Medium Effort(v1 快速初稿) 和 Max Effort(v2 极限投入) 两种模式下,纯粹通过编写 Python 脚本驱动 Blender 5.2 构建三个复杂度递增的 3D 场景(台灯静物、低多边形浮空岛、雨夜霓虹街道),并将最终成果烘焙导出到 Three.js 网页端进行实时 3D 交互对比。

我把这个仓库拉到本地完整拆解了一遍它的 Python 建场景脚本、GLB 烘焙管线、Three.js 预览器以及 Token 账单日志。这篇文章不聊虚的,直接从代码架构、视觉反馈闭环、WebGL 导出管线和 Token 成本结构四个维度,还原 AI 是如何“手搓”复杂 3D 工程的。

#realcodebear/blender-fable-5.2#Claude Fable 5.2#Blender 5.2#Three.js#GLB

Original · 10 min readRead the original
Group chat

No comments yet. Start the conversation.